First, lets summarise KAS's journey so far. For details, see the CHANGELOG.
Early releases, built over GTK.
A restart, replacing GTK with direct widget implementations including rendering
via the rgx crate. Initial theme traits, event-handling revision.
Lots of small but significant changes, including the introduction of the
Manager handle, user-defined configure code, scheduled updates (animation),
a scrollable region with scrollbars and addition of ToolkitProxy.
Only three weeks later, this delivered two new levels of draw API (portrayed by
clock and mandlebrot examples respectively), a "flat" theme, run-time theme
and colour-scheme switching, widget alignment, StretchPolicy, more adjustments
to event handling (removal of by-coordinate addressing), a RadioBox widget,
and window size limits.
This release revised widget traits into an integral family with opt-out
derivation (thus allowing future extension without breaking all existing
widgets), separating child-enumeration out of WidgetCore and separating
routing and handling of events.
Secondly, the release improved layout generation and notation, adding proper
margins, replacing Horizontal and Vertical with the four principal
directions, rewriting solve_seq to make only required changes (enabling
manual column resizing), and calculating the ideal window size before creation.
Thirdly (and no less significantly), the release added pop-up menus, including fixes for overlapping graphics and robust event handling (including delayed opening of sub-menus and navigation via tab and arrow keys).
Fourthly, it added access keys, including visible labels when Alt is depressed and locality to the visible pop-up layer.
Small additions included recursive disabled states for all widgets and an error
state for EditBox (set via a user-defined input guard). Widgets included a
Slider, a resizable Splitter, a ComboBox and menu widgets.
This release focussed on text presentation and editing with a new library, kas-text. The text editing experience has been massively improved, especially for multi-line text, and now supports bidirectional text, glyph shaping, and all the expected selection, navigation and editing functionality. Text presentation now supports justified text and faster re-wrapping when resizing the window.
Some tweaks to input handling and resizing landed, including (unconfigurable) short-cuts, double-click tracking and separate inner and outer margins.
Additionally, a CONTRIBUTING guide and this ROADMAP
have been added.
A continuation of the work on text, with intial support for rich text, exemplified via a Markdown parser. Underline and strike-through may be applied to texts at draw-time, and font adjustments (size, bold/italic and font family) may be applied as part of text layout.
To allow retrieval of data from temporary (dialog) windows, the
Window::on_drop method was added.
This release also simplifies distribution by bundling pre-compiled GLSL shaders.
This release focussed on one of the remaining hard problems: data sharing. A new framework for "view widgets" was introduced, allowing (interactive) views over shared data.
View widgets enable synchronised access to shared data from multiple locations. Additionally, by avoiding dedicated widgets for each data item in a set, views are scalable to large data sets.
Also included are several improvements to sizing, widget construction and event handling, as well as a new type-conversion library.
This release finally addressed one of the most obvious missing features of KAS:
images. At least, in their most basic form: static raster images loaded from
file sources. This new raster-image rendering system was then tweaked and used
as a replacement for the glyph_brush library, giving us better control over
glyph rendering and caching, and allowing a choice of font glyph rasterer.
The work on fonts and text did not stop there: as part of KAS-text 0.3, the
(rather large and mostly unused) dependency font-kit was replaced with
fontdb (plus a collection of custom aliases), extended to support font
fallbacks and run-time configuration, meanwhile rustybuzz was integrated as
a pure-Rust alternative to HarfBuzz for text shaping. These changes removed
all non-Rust dependencies from the text system.
Partially related to the above was the work on theme configuration, covering colour schemes, font size, font family (per font class and global), and glyph rastering.
Following the recent addition of image support, this release takes the
opportunity to revise the draw API, culminating in a new unified Draw trait
and DrawIface implementation, accessible via DrawHandle::draw_device. This
simplifies usage of mid-level draw APIs, e.g. in themes and the clock example.
Also in this release is SVG support, a Canvas widget, image buttons, and
fixed sub-pixel positioning (clearer rendering for small text).
This release responds to three key criticisms of KAS: (1) slow compile times, (2) non-standard keyboard navigation, (3) ugly themes. While these issues are not fully solved, they are significantly improved. Additionally, the latest version of WGPU now gives us a working OpenGL backend.
Widget trait revision: remove Handler::Msg associated type, replacing with a
variadic message stack: EventMgr::push_msg and try_pop_msg. Notify of data
model updates via broadcast.
New #[autoimpl] and impl_scope! macros, published in the
impl-tools crate.
Added a domain-specific macro-driven language for widget layout, thus permitting complex layout from a single widget.
Support sub-tree configuration. Make WidgetId a path.
Support theme- and widget-driven animations. Support momentum scrolling.
Stabilised support for Generic Associated Types, allowing some changes to data models.
Async rendering for Canvas and Svg widgets.
Support Linux's primary buffer. Support transparent and borderless windows with very basic support for a toolkit-drawn titlebar.
Removal of kas-theme crate (most contents merged into kas-core).
Add input data: widgets now have a Data associated type, passed by reference
to event handlers and to a new update method. The key advantages of this change are:
- Declarative specification of trees with changeable widgets. For example, before this change a
countermust construct a label displaying the initial count and explicitly update this label when the count changes; now the displayed label is effectively a function of the count. - As a direct result of the above, it is no longer necessary for parent nodes to refer to children by name, thus it is no longer necessary to use structs with named fields (at least in the common cases).
- "View widgets" are now much more tightly integrated with other widgets, making the
SingleViewwidget redundant.
For more on input data, read the design document.
These items are not versioned and appear only roughly in the expected order. Each should be a decently sized work item (roughly one release).
KAS has ad-hoc geometry types. Possibly it would be useful to use third-party types instead. See #95.
KAS supports pop-up menus since 0.4. Context menus are a little different, in that the widget opening the menu does not own it, yet can still customise it (enabling or disabling items like undo, paste, perhaps providing custom items, perhaps selecting a special context menu).
At the same time, the undo history should probably be removed from widgets and stored in some shared state.
Slightly related to this is support for global and standard shortcuts such as undo, copy selection, save file, quit app.
This is less a separate work item than it is a long-term goal, one which will require substantial help. KAS should attempt to discover initial configuration from the current desktop environment, e.g. the current font and font size, the icon set, locale and short-cuts.
Additionally, on Linux it is expected that the desktop environment provide a few standard dialog boxes (e.g. file open/save), not only for consistency but also security (e.g. a container may not let an app explore the filesystem). Such dialogs should automatically use desktop-provided equivalents where available.
Winit has at least partial support for IME now. Kas should add support for this
(mostly this means an additional Event variant, one or two API calls to
enable IME input for the current widget, and adjusting EditField).
AccessKit would appear to be the defacto Rust API for accessibility tools with
(limited) support from egui and Xilem toolkits. Widgets could support this via
additional code in the update method, possibly with additional tracking to
avoid unnecessarily replacing nodes.
Project Fluent offers the core functionality
required for localisation (l10n). We may need additional code for loading
resources and selecting languages; potentially also for fallbacks (in case of
incomplete translations). Widgets could translate their strings in the
configure method when fluent integration is enabled with labels treating
their input as a message key.
To investigate. Gstreamer integration should be viable when using a (new) OpenGL backend.
Integrating any video player as a child window should be possible (see Winit's
WindowBuilder::with_parent_window, which is not yet supported everywhere).
To investigate. Possibly use Servo (components), though it may not matter much.
The browser should probably be integrated via a child window (see Winit's WindowBuilder::with_parent_window, which is not yet supported everywhere).
We have plain text clipboard support via window_clipboard, but lack support for formatted text, images, etc.
It has been requested that winit support popup windows for things like menus which are not restricted to the parent window; currently it doesn't. See winit#950 (and other issues).
Winit does have support for this, but only in a very limited fashion. With its current event model it is difficult or impossible to determine the widget receiving a drop or under a hovered drop. See #98 and winit#1550.